The belief AI holds about Northwind
Reading Northwind's prior — the opinion ChatGPT, Perplexity, Gemini and Google AI already carry about you: the one adjective they're married to, the distance between what you can prove and what they believe, the source-types that built it, and how to move it. A ruler, not a verdict — what AI believes, and how to change its mind.
- The Prior at a glance — the one adjective AI is married to, the headline Evidence–Prior Gap, the vital signs
- The Association Map — Northwind vs each rival, the adjective AI marries to each
- The Evidence–Prior Gap — under-rated (evidence beats reputation) and over-rated (a decaying asset)
- Source Trace — the source-types feeding the belief
- The Playbook — the ranked source-types to win; which fixes auto-ship vs done-for-you
- Supporting context & Track over time — how the prior varies by engine and persona, Technical readiness (can AI read your site?) and Offerings readiness (can AI compare & recommend your product?); the belief as a health metric
- Appendices A–H — impact sizing, claim & source ledger, the question set, the 51 checks, method, statistics, glossary, sources index
Contents
Section numbers (§) and page references follow the printed edition of this report. Every entry is a live link in this on-screen edition — locked entries open after you unlock below.
What AI believes about Northwind — and how to move it
An AI assistant's answer about your brand isn't a lookup or a calculation — it's a prior: a belief assembled from thousands of pages the model read. This report reads Northwind's prior — the single adjective AI is married to you, the distance between what you can prove and what AI believes, the source-types that built the belief, and the ranked plays that move it. We measure the belief and trace where it came from; we do not adjudicate what is true.
We didn’t just count how often AI mentions you. We read 3,720 real AI answers and reconstructed the belief underneath them: the word AI reaches for when it describes Northwind, where that belief runs ahead of or behind what you can prove, and the exact source-types doing the work. Everything here is descriptive — you say X; AI believes Y; here is the gap, and how to close it.
The Prior at a glance — what AI believes about Northwind
The belief in one screen: the single adjective AI is married to Northwind, the headline Evidence–Prior Gap (where the belief runs ahead of or behind what you can prove), and the vital signs beneath it. Visibility is one of those vital signs — a thermometer, not the diagnosis. Everything is descriptive: we measure the distance between your evidence and AI's belief; we do not decide what is true.
Two things make this report trustworthy. First: we read the belief AI holds about you — the adjective it reaches for and the claims it repeats — not a raw mention count. Second: we never rule on what is true. The “evidence” side of every gap is what you told us you can prove; we simply measure the distance to what AI believes. AI also answers differently each time, so every percentage carries a ± margin, and gaps too small to be real are called ties.
The unit of analysis is the prior — the belief, not the mention. A mention count tells you whether AI talks about you; it cannot tell you what AI believes. So every one of the 3,720 captured answers was read for the belief underneath it — the adjective the engines reach for, and the recurring statements ("implementation takes 6+ months", "no SOC 2", "the affordable pick") — and each was traced to the source-type that produced it.
Evidence is yours; the gap is ours to measure. We do not adjudicate truth. Where this report reads a distance between what Northwind can prove and what AI believes (Part 3), the evidence side is brand-declared — captured at onboarding — and observable signals; the belief side is measured from the corpus. We report the distance: you say X; AI believes Y. Nothing here is a verdict on Northwind.
Every reading comes with its receipts. Four findings are heavy enough that we reproduce the captured answers as facsimile screenshots — Exhibits A–D in the Source Trace (Part 4) — each showing the exact prompt (with its question number from Appendix C), the engine, the capture date and run, the sentence that carries the belief highlighted, and, where the engine displayed citations, the actual clickable sources. Where a source is named anywhere in this report, it is linked; the full numbered index is Appendix H.
Every percentage carries uncertainty. Engines are stochastic (they answer differently each time). We ran every question three times per engine and report ± ranges (95% Wilson intervals, Appendix F). Differences smaller than the ~5-point noise floor (too small to be real) are treated as ties, and comparisons are tagged proven (certain beyond the margin) or indicative (directional). No forward projections — we describe the belief as it stands.
Ask any engine to describe Northwind and it hands you a single identity. Across 3,720 answers the association clusters tightly around one idea: a capable product that is a chore to adopt. This is the word that walks into the room before you do — and, today, it’s invisible to you.
Enterprise-heavy · slow to implement · hard to roll out. Elicited by asking the engines directly and aggregating across every answer, the association the models return for Northwind converges on a single frame: a good tool that costs you months to adopt. It precedes you into every shortlist and quietly sets the terms of the comparison.
The belief runs mostly behind what you can prove. On the attributes that decide deals, what you told us you can prove and what AI believes point in different directions. This is a ruler, not a verdict — you say X; AI believes Y — and the full two-sided read is Part 3.
| On this attribute | You say (brand-declared) | AI believes | Read |
|---|---|---|---|
| Implementation time | ~6 weeks median (cohort data) | "6+ months" | Under-rated |
| Security posture | SOC 2 Type II + ISO 27001, held today | "does not appear to hold SOC 2" | Under-rated |
| Integrations | 57 connectors, updated weekly | "about a dozen" | Under-rated |
| SMB price / value | priced at the premium end | "the cheap, easy pick" | Over-rated |
Six dials describe the health of the belief. Visibility (34/100) only says AI mentions you sometimes; it is the thermometer, not the diagnosis. The dials that matter more are about what AI believes and who hears it — the claim health, the tone, and the wild unevenness across personas.
Think of these six dials as a health chart, not a single grade. Visibility (34/100) just says AI mentions Northwind sometimes, rarely as a confident first pick. The dials that matter more are about perception: what AI actually says about you is driven by a handful of outside sources (Claim health 38 — the prevalence-weighted read on what AI asserts), the tone is mid-pack (Sentiment 52), and it's wildly uneven across your personas (Persona balance 22 — small businesses hear about you constantly, enterprises almost never). Your website itself is technically healthy (Technical 82); the problem is what's being said about you, and by whom.
You’ve seen the belief. The rest shows how to move it.
Unlock the full 30-page read — the Association Map (your word vs every rival's), the two-sided Evidence–Prior Gap, the Source Trace with captured screenshots, the ranked Playbook of source-types to win, plus the supporting depth and every appendix. Enter your work email to unlock.